Abstract

Economically motivated adulteration (EMA) of high-value geographical indication (GI) food ingredients poses serious risks to consumer trust and food safety. Rubus chingii Hu (RcH), a GI-designated edible-medicinal fruit originating from Dexing, is susceptible to adulteration with cheaper non-GI sources. This study developed a rapid, non-destructive near-infrared (NIR) spectroscopy approach combined with chemometric modeling for adulteration identification and ratio prediction of GI RcH. Two-dimensional correlation spectroscopy (2D-COS) using adulteration ratio as an external perturbation identified the 4200–6000 cm−1 region as the most sensitive spectral window, mainly reflecting changes in proteins, lipids, and carbohydrates. Using this 2D-COS-guided interval, we systematically compared four discriminant models (PLS-DA, SVM-DA, RF-DA, BPNN-DA) under ten preprocessing strategies. SNV + Smo-BPNN-DA achieved 100% test accuracy, while 1st D-RF-DA also showed excellent generalization (Ates = 99.67%). For adulteration ratio prediction, wavenumber selection (CARS, IWOA) significantly improved Raw-SVR performance (R2 from 59.8% to 78.6%) but provided no benefit—and sometimes degraded accuracy—for PLS, RF, and BPNN models, revealing a model-dependent effect rarely reported in food adulteration studies. The optimal regression model was SNV-BPNN-R (R2 = 96.96%); for computationally constrained scenarios, CARS + IWOA-2nd D-RF-R is recommended (R2 ≈ 95.93% with only 20–40 variables). Our findings provide a practical, physically interpretable, and transferable strategy for rapid EMA screening of powder-based GI food products, supporting on-site or online food authenticity control.

IPC Classification

Keywords